Mailbox math: optimizing sending volume without hurting deliverability

Once a set of mailboxes exists, the operational question becomes: how much can each one safely send *right now*, and when should that number increase. This is a different question from how many mailboxes you need in total - it's about the curve each individual mailbox follows as it earns more sending capacity.

The shape of a safe ramp curve

A typical gradual ramp follows something close to this pattern for a cold-outreach mailbox:

Week 1:  20/day
Week 2:  35/day   (+75%)
Week 3:  60/day   (+71%)
Week 4:  100/day  (+67%)
Week 5:  150/day  (+50%)
Week 6:  200/day  (+33%)
Week 7+: 250-300/day, held steady unless metrics support further growth

The percentage increase shrinks as the base grows - doubling volume in week 6 is a much larger absolute jump than doubling it in week 1, so the safe percentage increase naturally tapers even though the instinct is often to keep accelerating once early weeks go well.

The rule that actually governs the ramp: watch the signal, not the calendar

The schedule above is a starting template, not a fixed rule - the actual pace should respond to three live signals:

A mailbox that's hitting its week-4 milestone on the calendar but showing a rising bounce rate should not advance to the week-5 volume just because the schedule says so - the schedule is a default, the live metrics are the actual governor.

A worked example: when to deviate from the default curve

A mailbox at week 3 (60/day) shows complaint rate ticking up slightly after a list segment with uncertain opt-in quality gets mailed. Two options:

• Follow the calendar anyway → advance to 100/day in week 4 regardless → compounds the problem at higher volume.
• Hold at 60/day, investigate the segment, remove or fix it, confirm complaint rate normalizes → resume the ramp from week 3's volume once clean → costs a week of schedule but avoids a larger setback.

The second option is slower on paper and faster in practice, since a reputation setback from advancing on a bad signal typically costs more than a week to recover from.

What not to do

• Don't treat the ramp schedule as fixed once set - it's a starting assumption that live bounce, complaint and throttling data should override in either direction.
• Don't average volume across all your mailboxes and apply one number to all of them - a mailbox that's been sending cleanly for two months can safely carry more than one still in week 2, and treating them identically either under-uses the warmed one or over-pushes the new one.
• Don't resume full-speed advancement immediately after holding for a problem - resume from where you held, not from where the calendar says you'd be if nothing had happened.
• Don't confuse "the schedule says I can" with "the current metrics support it" - the schedule is the plan, the metrics are the actual permission.

Spreading volume across multiple mailboxes

The same total daily send looks very different depending on how it's distributed. Sending 300 messages/day from one mailbox puts all the reputation risk on a single sending identity; spreading that across five mailboxes at 60/day each means one mailbox having a bad day doesn't sink the whole campaign's deliverability, and each individual mailbox stays further from any single provider's per-account rate limits.

A simple way to think about total capacity:

Total safe daily volume ~= sum of each mailbox's current-week ramp ceiling
NOT: (number of mailboxes) x (one mailbox's ceiling)

That distinction matters because mailboxes added later start their own ramp from week 1, not from wherever the older mailboxes are. A domain with ten mailboxes - six at week 8 (300/day each) and four added last week (20/day each) - has a safe total closer to 1,880/day, not 3,000/day, until the newer four catch up.

Why domain-level signals override individual mailbox math

All of the above assumes the sending domain itself is healthy. If domain-level reputation is declining - rising spam-folder placement, a blocklist hit, or a worsening trend in Google Postmaster Tools - no individual mailbox's ramp math matters, because every mailbox on that domain inherits the same domain reputation. In that case the right move is to pause the ramp fleet-wide, diagnose the domain-level cause (often an authentication gap, a bad list segment shared across campaigns, or one mailbox's bad behavior bleeding into the shared domain signal), and only resume individual mailbox ramps once the domain-level trend has stabilized.

This is the practical reason volume math has to be read at two levels at once: per-mailbox, to decide whether *this* mailbox is ready for more, and per-domain, to decide whether the whole fleet should be advancing at all right now.

MailPilot runs exactly this kind of adaptive ramp automatically for every connected mailbox - increasing volume based on actual bounce, complaint and engagement signals rather than a fixed calendar, and holding or slowing the curve when the live data says to, across both OAuth and SMTP/IMAP connections.